We interpret industry discount rates as proxies for value investment, grounded in Q theory, while capturing gambling preferences through salience theory. Integrating these perspectives, we propose a novel asset pricing model (ST-ICAPM) that unifies value investment and salience factors, evaluating its pricing efficacy across Chinese industries from 2004 to 2023. Empirical results show that investment factors tend to negatively predict future returns, while growth factors command risk premia. However, profitability factors exhibit limited explanatory power. Investor expectations are primarily driven by profit growth, emphasizing the need to enhance profit stability for a value-oriented market. Salience intensity, especially when measured via eigenvector centrality within industry networks, serves as a strong negative predictor of returns, emphasizing the importance of conceptual connections over purely economic linkages in shaping investor behavior. Robust tests confirm that the ST-ICAPM outperforms benchmark models (FF3, CARHART4, FF5, ICAPM, and STCAPM) in terms of pricing power. Our findings emphasize the need to promote value investing and restrain gambling behavior as essential strategies to cultivate a resilient capital market in China.

投资者对宏观经济风险的评估如何影响资产价格一直是实证资产定价的难点之一。已有研究指出新闻文本会改变投资者对宏观经济的判断和预期进而影响股价，但如何有效提取与宏观经济风险相关的文本叙事信息来解释或预测资产价格变动，尚未达成共识。本文基于2007-2021年中国七家专业财经媒体的新闻报道数据，首次结合人工智能前沿领域的BERT大语言模型来测度新闻叙事注意力信息，然后利用稀疏工具主成分（Sparse IPCA）估计影响基本面的状态变量和影响资产价格的叙事定价因子。基于A股市场的实证检验发现：第一，本文利用新闻文本估计的状态变量对于消费、产出、国债收益率等宏观经济指标具有显著的预测效果，这表明新闻叙事蕴含着影响经济运行的信息。第二，相比CAPM、三因子等基准模型，基于新闻文本构建的叙事因子模型能更好地解释资产错误定价现象，并对未来资产价格的变化有更强的预测能力。第三，与基于关键词的文本分析方法（如LDA主题模型）相比，利用BERT提取文本信息可在保证因子模型简约性的基础上获得更优异的定价效果。本文的研究结论对于解释资产横截面收益差异有新的启示，同时为应用大语言模型于经济金融学研究抛砖引玉。